Copy-paste YAML templates and a complete worked example for documenting human roles.
Use this template to establish the boundaries and context of the role you're documenting.
AGENTIC REPLICA: [Role ID] role_title: [Official job title or functional name] department: [Where this role sits in the org] reports_to: [Who oversees this role] scope: [Brief description of what this role is responsible for] # Who/what does this role interact with? upstream_dependencies: - [Role/system that provides inputs to this role] - [Another source of inputs] downstream_consumers: - [Role/system that uses this role's outputs] - [Another consumer] primary_systems: - [Main software/tools this role uses daily]
List all distinct capabilities the person performs before documenting each one in detail.
skills_inventory: - id: SKILL-001 name: [Short name for this skill] frequency: [daily | weekly | monthly | on-demand | event-driven] time_spent: [Approximate % of role or hours/week] complexity: [low | medium | high] judgment_required: [low | medium | high] - id: SKILL-002 name: [Next skill...] # ...repeat for each skill
Tip: Most white-collar roles have 5-15 distinct skills. If you have fewer than 5, you may be grouping too broadly. If you have more than 15, you may be splitting too finely.
Complete this template for each skill in your inventory. This is the core work of Agentic Replica.
# ═══════════════════════════════════════════════════════════════ SKILL: [SKILL-001] name: [Descriptive name] # ═══════════════════════════════════════════════════════════════ # TRIGGER: What causes this skill to be activated? trigger: type: [message | schedule | event | request | data_change] source: [Where does the trigger come from?] frequency: [How often does this trigger occur?] examples: - "[Specific example of a trigger]" - "[Another example]" # INPUTS: What information/resources does this skill need? inputs: - name: [Input name] source: [Where it comes from] format: [email | document | database | API | verbal | etc.] required: [true | false] - name: [Another input...] # PROCESS: How is the work actually done? process: steps: - step: 1 action: [What they do] tools_used: [Software/systems used] time_typical: [How long this usually takes] - step: 2 action: [Next action...] # Tacit knowledge: The unwritten rules and judgment calls decision_points: - decision: "[What decision needs to be made?]" criteria: "[How do they decide? What factors matter?]" judgment_level: [routine | situational | expert] exceptions: - scenario: "[When things don't follow the normal path]" handling: "[What do they do instead?]" # TOOLS: What systems/resources are used? tools: - name: [Tool/system name] purpose: [What they use it for in this skill] access_level: [read | write | admin] - name: [Another tool...] # OUTPUTS: What does this skill produce? outputs: - name: [Output name] format: [report | email | database_update | decision | etc.] recipient: [Who receives this output?] sla: [Expected turnaround time] - name: [Another output...] # ESCALATION: When does this need human intervention? escalation: triggers: - "[Condition that requires escalation]" - "[Another escalation condition]" escalate_to: [Who handles escalations?] frequency: [How often do escalations occur? e.g., "~5% of cases"] # QUALITY: How do we know this was done well? quality_indicators: - [Metric or signal that indicates success] - [Another quality indicator] failure_modes: - [Common way this skill fails or produces poor results] - [Another failure mode]
After documenting all skills, identify patterns that apply across the entire role.
# ═══════════════════════════════════════════════════════════════ CROSS-CUTTING PATTERNS # ═══════════════════════════════════════════════════════════════ # Tools used across multiple skills common_tools: - tool: [Tool name] used_in_skills: [SKILL-001, SKILL-003, SKILL-005] access_needs: [What permissions would an agent need?] # Common escalation patterns escalation_patterns: - pattern: "[Description of common escalation type]" frequency: [How often across all skills] resolution: [How are these typically resolved?] # Dependencies between skills skill_dependencies: - "SKILL-002 often triggers SKILL-004" - "SKILL-001 output is input to SKILL-003" # Knowledge requirements domain_knowledge: - area: [Knowledge domain] depth: [basic | intermediate | expert] documentation: [Where is this knowledge documented, if anywhere?] # Time allocation summary time_distribution: high_frequency_skills: [Skills done daily] high_impact_skills: [Skills that matter most to outcomes] automation_candidates: [Skills with low judgment + high frequency] # Guardrails observed observed_guardrails: - "[Thing this person never does / always checks]" - "[Policy or rule they consistently follow]"
Here's a complete Agentic Replica for a Senior Financial Analyst role, showing how each template is filled out in practice.
AGENTIC REPLICA: FIN-ANALYST-001 role_title: Senior Financial Analyst department: FP&A (Financial Planning & Analysis) reports_to: Director of Finance scope: Prepares financial reports, analyzes variances, supports budgeting upstream_dependencies: - Accounting (GL data, month-end close) - Sales (pipeline forecasts, bookings) - Operations (headcount, cost inputs) downstream_consumers: - CFO (executive summaries) - Department heads (budget vs. actual) - Board (quarterly reporting package) primary_systems: - NetSuite (ERP) - Adaptive Insights (planning) - Excel (modeling) - Salesforce (pipeline data)
skills_inventory: - id: SKILL-001 name: Monthly Variance Analysis frequency: monthly time_spent: 25% complexity: high judgment_required: high - id: SKILL-002 name: Weekly Cash Flow Reporting frequency: weekly time_spent: 15% complexity: medium judgment_required: low - id: SKILL-003 name: Ad-hoc Analysis Requests frequency: on-demand time_spent: 20% complexity: variable judgment_required: high - id: SKILL-004 name: Budget Preparation Support frequency: quarterly (heavy in Q4) time_spent: 20% complexity: high judgment_required: medium - id: SKILL-005 name: Data Quality Monitoring frequency: daily time_spent: 10% complexity: low judgment_required: low - id: SKILL-006 name: Stakeholder Communication frequency: daily time_spent: 10% complexity: medium judgment_required: medium
# ═══════════════════════════════════════════════════════════════ SKILL: SKILL-001 name: Monthly Variance Analysis # ═══════════════════════════════════════════════════════════════ trigger: type: schedule + event source: Accounting month-end close completion frequency: Monthly, business day 5-7 examples: - "Email from Controller: 'January books are closed'" - "Slack notification from #accounting: 'GL ready for FP&A'" inputs: - name: General Ledger Export source: NetSuite format: CSV export, ~15,000 rows required: true - name: Budget File source: Adaptive Insights format: Excel workbook required: true - name: Prior Month Actuals source: Internal SharePoint format: Last month's variance report required: true - name: Department Context source: Slack conversations, meeting notes format: Unstructured required: false (helpful for commentary) process: steps: - step: 1 action: Export GL data from NetSuite using saved search "FPA Monthly" tools_used: NetSuite time_typical: 10 minutes - step: 2 action: Import GL into variance template, run pivot refresh tools_used: Excel time_typical: 15 minutes - step: 3 action: Identify variances > $10K or > 10% from budget tools_used: Excel (conditional formatting flags these) time_typical: 20 minutes - step: 4 action: Research root causes for flagged variances tools_used: NetSuite drill-down, Slack, Email time_typical: 2-4 hours (varies by month) - step: 5 action: Write variance commentary for each flagged item tools_used: Word/Excel time_typical: 1-2 hours - step: 6 action: Send draft to Director for review tools_used: Email time_typical: 5 minutes decision_points: - decision: "Which variances need investigation vs. known/expected?" criteria: "Known items (e.g., approved overspend) are noted but not investigated. Unexpected items get full root cause." judgment_level: situational - decision: "How much detail in commentary?" criteria: "More detail for items CFO will ask about. Less for operational noise." judgment_level: expert - decision: "Flag as concerning or just informational?" criteria: "Concerning = trend likely to continue, or one-time but significant. Informational = isolated, expected to normalize." judgment_level: expert exceptions: - scenario: "GL data has obvious errors (negative revenue, missing cost centers)" handling: "Stop analysis, email Controller immediately, document in report" - scenario: "Can't reach department head for explanation" handling: "Note as 'pending explanation' and follow up next day" tools: - name: NetSuite purpose: GL export, transaction drill-down access_level: read (financial reports role) - name: Excel purpose: Variance calculations, pivot tables, formatting access_level: local application - name: Adaptive Insights purpose: Budget data retrieval access_level: read (analyst role) - name: Slack purpose: Quick questions to department heads access_level: member of finance, department channels outputs: - name: Monthly Variance Report format: Excel workbook + PDF summary recipient: Director of Finance, CFO sla: Business day 8 of month - name: Department Variance Summaries format: Email with attached Excel slice recipient: Individual department heads sla: Business day 10 of month escalation: triggers: - "Variance > $100K with unknown cause" - "Potential fraud indicators (duplicate payments, unusual vendors)" - "Data quality issues blocking analysis" escalate_to: Director of Finance frequency: ~2-3 times per year quality_indicators: - Report delivered by SLA (measured) - CFO has no follow-up questions on documented items (feedback) - Variances correctly categorized (audit check) failure_modes: - Missing a material variance that CFO discovers in board meeting - Commentary that contradicts department head's understanding - Late delivery delaying executive review cycle
# ═══════════════════════════════════════════════════════════════ CROSS-CUTTING PATTERNS # ═══════════════════════════════════════════════════════════════ common_tools: - tool: NetSuite used_in_skills: [SKILL-001, SKILL-002, SKILL-003, SKILL-005] access_needs: Read access to GL, AP, AR modules; saved search execution - tool: Excel used_in_skills: [All skills] access_needs: Full Excel capabilities, macro execution - tool: Slack used_in_skills: [SKILL-001, SKILL-003, SKILL-006] access_needs: Message in #finance, DM to department heads escalation_patterns: - pattern: "Data quality issues in source systems" frequency: Monthly (during close) resolution: Controller fixes in NetSuite, analyst re-runs - pattern: "Executive requests conflicting with existing priorities" frequency: ~2x per month resolution: Director triages and re-prioritizes skill_dependencies: - "SKILL-001 completion triggers SKILL-006 (distribute to stakeholders)" - "SKILL-005 issues can block SKILL-001 and SKILL-002" - "SKILL-003 often informs SKILL-004 assumptions" domain_knowledge: - area: GAAP accounting principles depth: intermediate documentation: Company accounting policy manual - area: Company cost structure and business model depth: expert documentation: Tacit / learned on job - area: NetSuite navigation and reporting depth: expert documentation: Internal wiki + vendor docs time_distribution: high_frequency_skills: [SKILL-005, SKILL-006] high_impact_skills: [SKILL-001, SKILL-004] automation_candidates: [SKILL-002, SKILL-005 - low judgment, high frequency] observed_guardrails: - "Never share unreviewed numbers with non-finance stakeholders" - "Always verify GL totals match trial balance before analysis" - "Never discuss board-level financials outside exec team" - "Always CC Director on communications to C-suite"